Steerer Model

• Goal: make RL for thousands of environments unnecessary.
  • Train a small steerer model to guide a good base coding-model LLM.